pixi-install-nvidia
Use when the user says "use pixi to install <some nvidia tool>" (or similar) and wants NVIDIA/CUDA/GPU packages installed via Pixi (no sudo/apt), e.g., CUDA toolkit pieces, cuDNN/NCCL, PyTorch CUDA builds, RAPIDS.
Best use case
pixi-install-nvidia is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Use when the user says "use pixi to install <some nvidia tool>" (or similar) and wants NVIDIA/CUDA/GPU packages installed via Pixi (no sudo/apt), e.g., CUDA toolkit pieces, cuDNN/NCCL, PyTorch CUDA builds, RAPIDS.
Teams using pixi-install-nvidia should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/pixi-install-nvidia/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How pixi-install-nvidia Compares
| Feature / Agent | pixi-install-nvidia | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Use when the user says "use pixi to install <some nvidia tool>" (or similar) and wants NVIDIA/CUDA/GPU packages installed via Pixi (no sudo/apt), e.g., CUDA toolkit pieces, cuDNN/NCCL, PyTorch CUDA builds, RAPIDS.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Pixi Install NVIDIA ## Trigger Use this skill when the user asks to install NVIDIA tooling *via Pixi*, especially in the form: - "use pixi to install <some nvidia tool>" Examples: - "use pixi to install cuda" - "use pixi to install nvcc" - "use pixi to install cudnn/nccl" - "use pixi to install pytorch cuda 12.1" ## Overview This skill provides workflows for setting up **No-Sudo**, **User-Space** GPU environments. By installing CUDA toolkits and libraries via Pixi, you avoid modifying the host system (`apt install`) and ensure perfect reproducibility across different machines. ## Key Benefits * **No Root Needed**: Install compilers (NVCC) and libraries (cuDNN, NCCL) without sudo. * **Isolation**: Project CUDA version is independent of the host's `/usr/local/cuda`. * **Reproducibility**: `pixi.lock` guarantees the exact same driver-compatible libraries everywhere. ## Workflow ### 1. Channel Configuration Ensure the correct channels are present in `pyproject.toml`. The order determines priority. * **Command**: `pixi project channel add nvidia` (and `pytorch` if needed). * **Priority**: `nvidia` **MUST** be prioritized over `conda-forge` for all NVIDIA tools and libraries (CUDA, cuDNN, NCCL, etc.), unless the user explicitly requests otherwise. ### 2. Adding GPU Packages Use specific versions to ensure compatibility between CUDA and the framework. **Always** prefer the `nvidia` channel for these packages. #### PyTorch (Recommended) ```bash pixi add pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia ``` #### CUDA Toolkit Components Instead of the massive `cuda-toolkit`, consider adding only what's needed: ```bash pixi add cuda-compiler cuda-libraries-dev ``` ### 3. Verification After installation, verify GPU visibility: ```bash pixi run python -c "import torch; print(torch.cuda.is_available())" ``` ## Troubleshooting & References * **CUDA Version Mismatch**: Check `nvidia-smi` on the host to ensure the installed `pytorch-cuda` version is supported by the host driver. * **Library Loading Issues**: If `libcuda.so` or `libcudart.so` are not found, ensure the environment is activated (`pixi shell`). * **Detailed Package List**: See [nvidia-packages.md](references/nvidia-packages.md) for a comprehensive list of available NVIDIA and GPU libraries. * **Compiling Code**: See [compiling-cuda.md](references/compiling-cuda.md) for instructions on using `nvcc` and `cmake` with the Pixi-managed toolkit. ## Resources ### references/ * **[nvidia-packages.md](references/nvidia-packages.md)**: Comprehensive guide on channels, packages, and versioning for NVIDIA ecosystems. * **[compiling-cuda.md](references/compiling-cuda.md)**: Guide for compiling CUDA C++ code using the Pixi-managed toolchain.
Related Skills
pixi-make-offline-channel
Use when the user wants to create a self-hosted, offline-installable Conda channel (mirror) containing a specific subset of packages using Pixi.
pixi-make-cu-build-env
Guides the agent to setup a new or existing Pixi environment for compiling C++ and CUDA code. It ensures the correct compilers, toolkits, and CMake configurations are in place for a robust user-space build.
init-pixi-project
Initialize and scaffold a new Pixi-managed Python project, or upgrade an existing one. Use when the user explicitly mentions "init", "initialize", or "setup" in conjunction with "pixi". Handles existing projects by prompting for confirmation before merging.
codex-cli-install
A three-part Codex CLI setup skill with subskills for installation, skip-login configuration, and custom API-key launcher setup. Use when the user wants Codex CLI installed or configured on the current host, with instructions adapted to the actual operating system and runtime environment.
claude-code-install
A three-part Claude Code setup skill with subskills for installation, skip-login configuration, and custom API-key launcher setup. Use when the user wants Claude Code installed or configured on the current host, with instructions adapted to the actual operating system and runtime environment.
vscode-make-offline-installer
Build a VS Code air-gapped "offline kit" for Remote-SSH: desktop VS Code installers/archives for Windows/macOS/Linux, matching VS Code Server+CLI tarballs for headless Linux (commit-aligned), and pinned extension .vsix bundles for both local (client) and remote (server) sides. Use when you need a reproducible offline deployment, or when you want to mirror the VS Code release+extensions currently installed on a host.
docker-pixi-project-offline
Scripted process to build and verify a Pixi-managed project in Docker, then produce a portable WORKDIR/product directory for air-gapped use.
pei-docker-usage
Helper for PeiDocker (`pei-docker-cli`). Trigger ONLY when the user explicitly requests PeiDocker usage OR when working within a PeiDocker-generated project (indicated by `user_config.yml`).
conan-basic-usage
Basic operations for the Conan C++ package manager. Use when the user explicitly asks to 'use conan' for tasks like creating projects, installing dependencies, or building packages, or asks for 'how to' guidance on Conan setup.
explore-dnn-model
Manual invocation only; use only when the user explicitly requests `explore-dnn-model` by name. Explore how to run a given DNN model checkpoint in the current Python environment by locating weights + upstream source code, resolving dependencies with user confirmation, running reproducible experiments under `tmp/`, and producing reports about I/O contracts, timing, and profiling.
openspec-ext-revise-by-decision
Manual invocation only; use only when the user explicitly requests `openspec-ext-revise-by-decision` by exact name. Revise OpenSpec change artifacts from a review or decision document that contains questions plus `DECISION` blocks, applying chosen decisions from a review file such as `openspec/changes/<change>/review/review-*.md` back into proposal, design, specs, and tasks.
openspec-ext-review-plan
Review an OpenSpec change (or a single OpenSpec change artifact file) for completeness, coherence, and alignment with existing system design; capture actionable feedback plus open questions; write a review report under the change directory (review/review-YYYYMMDD-HHMMSS.md).